MSMTRIU-Net: Deep Learning-Based Method for Identifying Rice Cultivation Areas Using Multi-Source and Multi-Temporal Remote Sensing Images | JoVE Visualize

Area of Science:

Agricultural remote sensing
Geospatial analysis
Deep learning applications

Background:

Timely and accurate identification of rice cultivation areas is vital for agricultural policy and understanding crop distribution.
Traditional remote sensing methods using single-source or single-temporal data limit the utilization of rice information across different growth stages and image types.
Existing methods often yield unsatisfactory identification results due to incomplete data utilization.

Purpose of the Study:

To develop and evaluate a novel deep learning method for identifying rice cultivation areas using multi-source and multi-temporal remote sensing data.
To leverage both optical (Landsat-8) and radar (Sentinel-1 PolSAR) data to capture diverse spectral and scattering traits of rice.
To enhance the accuracy and continuity of rice area mapping.

Main Methods:

A U-Net based deep learning model, named Multi-Source and Multi-Temporal Rice Identification U-Net (MSMTRIU-NET), was developed.
The model integrates multi-temporal features from Landsat-8 optical and Sentinel-1 Polarimetric Synthetic Aperture Radar (PolSAR) datasets.
Training involved feeding spectral reflectance and polarimetric scattering traits of rice across different periods into the network.

Main Results:

The MSMTRIU-NET model achieved high classification precisions in identifying rice cultivation areas on China's Sanjiang Plain.
Incorporating multi-source and multi-temporal image information demonstrably improved classification performance.
The resulting classification maps showed enhanced continuity and more accurate delineation of rice regions.

Conclusions:

The proposed MSMTRIU-NET method effectively identifies rice cultivation areas with high accuracy.
Utilizing multi-source and multi-temporal remote sensing data significantly boosts the performance of rice area identification.
This approach provides a more continuous and accurate representation of rice distribution, aiding agricultural monitoring and policy-making.